Hi Praveen,

You might have a look at the correlations between the scour depth data and the parameters. If there is a high correlation between the scour depth and a particular parameter in the data, you would expect that this high correlation is also present in a (good) model and hence the model output (scour depth) is sensitive to variations in this parameter.

Good luck,

Martijn

I think without model its very difficult to identify the sensitivity parameter.

Its required simulation for any equation.

But you can do

t-state and confidence level analysis test by changing parameter which give scour. I'm not expert in this field.

Its better to use any model and compare your scour observed results and then you can find better simulation coefficient which show

Article Scour effects in piled structures A sensitivity analysis

Thank You Leelambar Singh for your suggestion

im looking for the sensitivity analysis on the Dependent parameter over independent parameter so that i can avoid the selection of less sensitive parameter. After modelling and doing sensitivity analysis would help in model calibration and computational time. looking for purely the statistical technique to carry out sensitivity, gone through already Sobol, variance decomposition, regression based techniques, Taylor series expansion of the explicit function, these technique area really good but not for my purpose to fulfill.

looking forward for the fruitful suggestions over aforesaid techniques if these are helpful

Thank you.

Hi Praveen,

You might have a look at the correlations between the scour depth data and the parameters. If there is a high correlation between the scour depth and a particular parameter in the data, you would expect that this high correlation is also present in a (good) model and hence the model output (scour depth) is sensitive to variations in this parameter.

Good luck,

Martijn

Resapected Martijn J. Booij Sir,

Thank you for the suggestion

Dear Praveen,

The following article might be useful:

1) Article Prediction of current-induced local scour around complex pie...

2) Article Scour hole depth prediction around pile groups: review, comp...

Sincerely,

Behshad

Have you tried the analysis of variance (ANOVA)? You mentioned you tried variance decomposition but that this does not work for your purposes... I think with ANOVA you can look at which factors have statistically significant effects (e.g., p < 0.05) on the scouring, and also look at the size of the effects of independent factors and their interactions. These p-values and effect magnitudes can inform you of which of the factors are more sensitive. A similar analysis can be done with a regression approach. I usually carry out these multi-factor analyses on Minitab (http://www.minitab.com). For more information on these types of techniques, you can consult Montgomery's text book: Montgomery, Douglas C. Design and analysis of experiments. John wiley & sons, 2017.

Praveen---other stat techniques:

Random Forest

Relative Importance

Step-wise regression

Boruta method

Information value and weight of evidence

link: http://rstatistics.net/variable-importance-of-predictors-that-contribute-most-significantly-to-a-response-variable-in-r/

Also; Principle Components Analysis (a regression technique)

One or more of these techniques should prove useful

Thank you Respected Felipe Hernández and James C. Trask for the valuable answer.

Hi Praveen,

I would recoommend the same as Dr. Martijn J. Booij , looking at scatter plots between the single parameters should give you a globally good perception of the sensitivity. However if you have a decent amount of data you could also estimate some variance based indices to see have it in numbers and to figure whether there is a correlation between input parameters playing a role or not.

Let me know if you would like some references to read about it!

Friendly Regards,

Mauro

Praveen Rathod : please see whether this article might be of any help.

Hi Preveen;

To do sensitivity analysis you may have 2 options:

1- Local sensitivity analysis: Based on the parameters in your developed model. This will clarify the importance of various parameters contributing in your final model.

2- The global sensitivity analysis: Based on the data set. Here, your want to choose the most important parameters to use them to evolve a model.

For more information, you can refer to my papers specially the following one:

Article Prediction of non-breaking wave induced scour depth at the t...

Best wishes

Since your question relates to the global sensitivity analysis, I recommend the following procedure:

1- Develop various models using the various input variables (in different models use various inputs, e.g. if you have n input parameters, in 1st mode all the inputs; in 2nd model use the (n-1) input parameters, etc. ).

2- Compare their performance based on the statistical parameters.

based on these information, you will be able to select the most important parameters contributing the model.

Keep in mind that, since soft computing models are based on the data points (they do not consider the physics of the problem), sometimes one input parameter looks unimportant but it plays an important role physically, so you can not neglect it in your modelling procedure.

I would also recommend to basically apply the solution proposed by Dr. Martijn J. Booij, if you have access to relevant data. It would be better if you can train an ANN model on the available data and then apply the ANN model to evaluate the sensitivity of results to different parameters.